South Africa has prioritised large-scale measurement and monitoring of reading comprehension outcomes across the country in standardised assessments. Thus, South African learners have been participating in standardised assessments such as the Progress in International Reading Literacy Studies (PIRLS) (Howie et al. 2008) and the Southern and Eastern Africa Consortium for Monitoring Educational Quality (Moloi & Chetty n.d.). According to Howie et al. (2008), the prioritisation of these measurements was because the Department of Education has recognised language and reading literacy as some of the important priorities in education. Participation in these assessments has highlighted issues relating to South African learners’ reading capabilities which are concerning and, therefore, warrant scrutiny to understand where the problem lies. The results of these assessments, particularly those of the PIRLS, show that most South African Grade 4 learners are unable to comprehend what they are reading.

The four PIRLS studies that South African learners participated in, that is, the 2006, 2011, 2016, with the latest being the 2021 cycle, show that South African Grade 4 learners are unable to read with comprehension. The 2021 results even show that gains made in the past 10 years were reversed because of learning losses incurred during the coronavirus disease of 2020 (COVID-19) period (Ardington, Wills & Kotze 2021). The latest study showed that 81% of South African learners could not read for meaning in any language, obtaining a score of 288 (SE = 4.4) points; and that Sesotho learners’ performance plunged from 319 in 2016 to 258 points in 2021 (Department of Basic Education 2023). This is a steep fall, which is worrying. The 258 points represent the PIRLS benchmarks that categorise learners’ reading ability numerically from the lowest (400) to the highest (650) points within a scale of 0–1000 (Phillips 2014). These benchmarks or levels, as they will also be referred to, indicate comprehension processes in which learners understand or do not understand, with the levels being the ability to:

These authors further explain the level of text engagement that readers should display at each level. However, because Grade 4 Sesotho learners achieved way less than 400 points, which indicates the inability to retrieve explicitly stated information or, in other words, inability to demonstrate literal comprehension and making straightforward inferences, the article shall focus on what retrieving explicitly stated information means to clarify South African learners’ reading level in general and Sesotho learners in particular. However, it is important to mention that the sampled Grade 4 learners who wrote PIRLS 2021 include both learners to whom Sesotho is a home language and some who may not necessarily speak the language at home but learn it at school because of language dynamics in the community and thereby in schools.

The last two benchmarks (interpret and integrate ideas and information, and evaluate and critique content and textual elements) require readers to make inferences at a global level of text, which is unattainable if the first two prove to be challenging. To make inferences at the third and fourth benchmarks, readers need to first access the literal meaning of texts. Mullis and Martin (2019:13) explain that to retrieve explicitly stated information, readers will engage with texts to:

These authors further mention that to retrieve this information successfully, the reader needs ‘immediate or automatic’ understanding of the words, phrases or sentences while being able to recognise that they are relevant to the information sought. Before a reader can understand words, a they needs to recognise or decode them. For example, a sight word such as ‘above’ will be recognisable if the reader has encountered it before and knows its meaning. However, a regular word such as ‘bold’ can be recognised if a reader had encountered it multiple times before or decoded it, if it is new. Sesotho words are all regular and thus need to be decoded to be recognisable. However, words that have been read repeatedly can become like sight words and be automatically recognisable; these can include function words which in Sesotho are short, single-letter or single-syllable words such as ‘e’, ‘a’ and ‘le’ or ‘ho’.

The inability to answer questions that require readers to retrieve explicitly stated information in texts, therefore, may indicate that readers are unable to recognise or decode words successfully. Magliano et al. (2023) state that even at secondary school level, reading challenges often revolve around decoding, fluency, vocabulary and comprehension and these are skills that are identified as important by the National Reading Panel of 2000.

This study focuses on decoding, which is espoused in the knowledge of letter sounds and fluency as basic skills that should support reading with comprehension. Pretorius and Klapwijk (2016) state that without decoding, comprehension is not possible. Decoding is one of the two skills advocated by Gough and Tunmer’s (1986) ‘simple view of reading’ (SVR).

Reading comprehension is said to involve visually perceiving letters and words, converting them into speech sounds (decoding), parsing linguistic information, developing an initial understanding of propositions based on the given information, and integrating propositions for an accurate and coherent mental representation, referred to as the process of making meaning from written text (Young-Suk 2023:335). Because of its complex nature, reading comprehension problems can thus stem from a deficiency of any of these skills (Oakhill, Cane & Elbro 2019).

Gough and Tunmer’s (1986) SVR model simplifies this otherwise complicated process, while the active view of reading (AVR) model by Duke and Cartwright (2021) shows reading in its comprehensive form. This latter view of reading is an extension of the simple view of reading (SVR) and thus both models will be discussed as they fit as lenses for analysing reading in important ways.

The SVR is an interactive reading model that proposes that to read with comprehension, readers should be able to decode written language and have linguistic comprehension (Gough & Tunmer 1986). Linguistic comprehension, though important, is outside the scope of this study. The SVR model analogises reading comprehension as the product of a multiplication sum exemplifying that the two skills, decoding multiplied by linguistic comprehension, equal reading comprehension (D × L = RC). The multiplication sign represents the point at which the two skills interact to materialise comprehension. The model thus predicts that a proficient reader at any level will possess both strong decoding and language comprehension skills, while a poor reader will be lacking one or both of these competencies (Wills et al. 2021). Many researchers have utilised the SVR model as an analytic framework for assessing reading comprehension (Protopapas et al. 2012). Hogan, Adlof and Alonzo (2014:199) state that the model ‘does an excellent job in explaining the reading comprehension developmental span not only in beginner readers but also in adult readers’. An investigation by Kendeou, Savage and Van den Broek (2008:34) found that the ‘SVR described the performance of university students with and without known reading difficulties’. Florit and Cain (2011) studied the developmental patterns of relations between reading comprehension, decoding, and linguistic comprehension in children reading in English and those reading transparent languages and found that SVR accounts for approximately 40% – 80% of the variance in reading comprehension for readers ranging from 8 to 16 years. This means there were differences in how decoding and linguistic comprehension played a role in the reading comprehension of the said sample. Languages with a transparent or shallow orthography e.g. IsiZulu are writing systems with direct spelling sound correspondences, while opaque languages like English have inconsistent relationships between letters and sounds. Though the SVR model continues to be a helpful lens for teachers and researchers for reading instruction and analysis, it has also been viewed as limited in capturing what is entailed in this complex skill (Catts 2018). Duke and Cartwright’s (2021) AVR model broadens the SVR model by first using the broader term word recognition and language comprehension instead of decoding and linguistic comprehension. In this model, the authors (ibid.) show that reading is facilitated by three major skills, viz, active self-regulation (motivation, engagement and executive function skills), word recognition (phonological awareness, alphabetic principle, phonics knowledge and decoding or recognition of words at sight) and language comprehension which includes (cultural and other content knowledge, reading specific background knowledge, verbal reasoning, language structure and theory of mind). Another component of the AVR model are the skills that capture the overlap between word recognition and language comprehension, which Duke and Cartwright (2021) conceptualise as the bridging processes that will be briefly defined later. The concepts of interest that will be further explored in this article are decoding and reading fluency, the latter being a skill that is viewed as the bridge between decoding and comprehension (Pikulski & Chard 2005).

Normally distributed data have a bell-shaped curve described by its mean, standard deviation and extreme values (Mishra et al. 2019). To test for normality, a Shapiro–Wilk test was used. Table 1 shows the results of this test.

The results of the test reveal that the data departed significantly from normality for the three variables tested. The p-values for all three were less than 0.05. These results may be an indication that there may be a relationship among the variables tested whose means are all deviating from the norm. Table 1 shows the normality test results for simple and complex letter-sounds as well as wcpm.

Table 2 shows the results for all skills tested. Overall, the mean scores reveal that most learners achieved scores around 50% or less of total scores for each of the variables.

This study aimed to assess Sesotho Grade 4 learners’ knowledge of simple and complex letter sounds and oral reading fluency to establish their relationship with oral reading comprehension. Overall, the scores showed low performance across all measured variables.

The CAPS document suggests the synthetic phonics as an approach for teaching reading. As explained in the literature review, the synthetic approach requires that the correspondence of letters and their sounds be taught for readers to blend them to decode words in texts (Johnston et al. 2011). To reach decoding levels that will positively impact comprehension, Setswana reading benchmarks and thresholds require learners to correctly sound 40 letters by the end of Grade 1 (Wills et al. 2021). Poor letter sound knowledge, therefore, translates to poor decoding (Nieto 2005), and poor decoding precipitates dysfluent reading, which impacts reading comprehension negatively (Ganhdi et al. 2017).

The first research question sought to find out ‘What is the overall Grade 4 Sesotho learners’ performance on measures of letter sound knowledge (simple and complex letter sounds)?’ The outcomes showed that most learners could identify only 32.05 of simple letters sounds and above just 20.81 of complex letter sounds.

These results are consistent with those of other studies that measured letter sound variables in other African languages from both the Sotho and Nguni clusters. A study by Spaull et al. (2020) of Northern SeSotho, Xitsonga and isiZulu Grade 3 learners also revealed that learners could only read 28 letters correctly per minute. Another study by Ardington et al. (2021:13), which analysed isiXhosa, isiZulu and siSwati Grade 3 learners, indicated that many learners ‘are not mastering this basic skill by the end of Grade 3’. One in 10 learners across the pooled data ‘was unable to sound one letter correctly at the end of the Foundation Phase’.

Additionally the results revealed that learners made a variety of errors in sounding out both sets of letter sounds such as:

The errors identified earlier indicate that the participants have not mastered this basic reading skill considered prerequisite for decoding according to the SVR (Gough & Tunmer 1986). Poor performance in this foundational, constraint skill, which Durgunoglu and Ӧney (1999) say can be grasped by the end of Grade 1 for transparent languages, could be attributable to a variety of factors, including:

The participants’ deficient knowledge of letter sounds manifested itself in their reading fluency.

Reading with fluency is considered the bridge between decoding and comprehension (Pikulski & Chard 2005). Without it, reading effort is directed towards decoding words rather than comprehension them (Ehri 2005). The oral reading fluency results for this study were in relation to the research question, ‘How many words per minute can Sesotho Grade 4 learners read correctly per minute?’ Statistics showed that the mean score was 34.32 wcpm, which constitutes just above half the 60 wcpm requirement by the end of Grade 3 stipulated by Setswana reading benchmarks. Low performance in oral reading fluency was also found in studies by Ardington et al. (2021), Spaull et al. (2020) and Wills et al. (2021).

During this test, reading errors on short function words with single letter and single syllable such as e, o, and ke, sa and lexical words with both simple and complex sounds, such as haholo [a lot] and leeba [dove] and those with digraphs and trigraphs such as tsatsi [day] and tshoswana [ant], were mispronounced. Learners were not sure about initial and end sounds in words which required them to re-read them in a quest to correct errors. Reading errors on short function words was puzzling as these are words that frequently appear in all Sesotho texts and learners should have automatised. Within the 34-word range that most learners could read, there were 17 of these words, almost half the number of words that learners could read within a minute. The question is, what type of errors learners make on short Sesotho function words, and what impact does it have for reading fluency? This is a puzzle that can be solved by further research.

Coming back to reading errors on lexical words, the two most mispronounced words in this test were tshoswana [ant] and tjodietsa [singing of birds] by 51% and 35% of participants, respectively. The word tshoswana appeared twice within the 34-word range, potentially increasing the number of incidences of mispronunciation as some learners errored the second time. Of the 51% participants who mispronounced the word the first time, 21% still mispronounced it the second time. However, the fact that only 21% of the participants mispronounced the word the second time shows that 30% of the participants may have read it correctly because of re-reading it, highlighting the importance of repeated reading. In the letter sound test, the phoneme /tsh/ contained in the word tshoswana was correctly pronounced by only 45% of the participants. In the letter sound test, the grapheme (tsh) was also inconsistently correctly pronounced by learners at the various intervals that it appeared in the test as it appeared more than once on the letter sound chart. The grapheme tsh was confused with either ts or tjh, while diagraph tj was pronounced either as j, tjh or as ts.

Tjodietsa, the second most mispronounced word, was the 36th word in the oral reading text, meaning that the 35% of learners who mispronounced it are outside the bracket of the 34.32 wcpm readers. Perhaps the percentage of learners who mispronounced the word could have been higher had learners read up to the 36th word, increasing the mean number of words read incorrectly per minute.

An interesting observation in relation to words with complex letters was the reading error the sampled learners made on words ending with digraph ng. The digraph appeared at the end of words such as moriting [in the shade], hobaneng [why], nokeng [at the river] and leng [other]. The phoneme proved challenging as most learners re-read some of the words containing it having initially omitted the sound only to go back to the beginning of the word to sound it out. Or maybe because words ending with ‘ng’ are odd to the syllable structure of consonant-vowel cv which most Sesotho follow. Paige (2020) also mentions that students with an inadequate inventory of words recognised automatically are unlikely to become fluent readers, and dysfluent readers are unlikely to comprehend what they read. ‘It could also be that’ these learners have rarely encounters words ending wing ‘ng’ in their inventory. The slow reading pace of participants precipitated by inaccurate reading of words because of a deficient letter sound knowledge could be attributable to lack of reading practice. Stanovich (1986) states that differences in reading are partially because of the amount of practice they receive in reading and writing. These letter sound knowledge and reading fluency results described earlier do not bode well for comprehension. Slocum et al. (1995) state that students who score higher on comprehension tests were also faster in reading letters and words in text. Ganhdi et al. (2017:1) say that: ‘Students who have challenges decoding text, their word reading accuracy interferes with measuring their comprehension and understanding of text’.

Oral reading comprehension results of a 2.47 mean score out of 6 calculated to 41% illustrated low comprehension as predicted. The comprehension test consisted of three literal and three inferential questions. Statistics show that, generally, the percentages of learners who responded correctly to literal questions hovered around 50%, while the numbers of correct responses plummeted for inferential questions. Participants who could not answer inferential questions 4 calculated to 69% while for question 6 they made up 78%. The oral reading comprehension results of Spaull et al.’s (2020) study revealed that learners achieved 49%, which is 8% more than for this study. These results are also comparable to those for Setswana EGRA (Wills et al. 2021).

Poor performance on literal and straightforward inference questions, which require readers to search for an answer by referring to the text (Eason et al. 2012) from a short text, potentially points to a struggle with decoding and recognising words as proved by the oral reading fluency results of this study. It could indicate that learners were unable to read both the questions and the corresponding text successfully. The chain reaction from letter sound knowledge, to wcpm and to comprehension shows how these skills are all essential contributory factors that can make or break reading with comprehension.

The last research question that sought to establish relationships among the three measured variables asked: What is the relationship between letter sound knowledge, oral reading fluency and oral reading comprehension of Grade 4 learners? The Spearman correlation test results showed that there were significant correlations between all variables tested. The study by Vaz (2024) illustrated not only how letter sounds affect oral reading fluency and how fluency affects comprehension, but also how letter sounds have a direct influence on comprehension. Figure 1 shows these relationships.

Other studies that also investigated relationships among decoding, reading fluency and text comprehension show strong correlations (Ardington et al. 2020; Spaull et al. 2020; Wills et al. 2021). The strong correlation among these variables may indicate that there are relationships among them and all play a role to achieving comprehension, and that a weak grasp of each may be the weak link that contributes to poor comprehension.

A conclusion that can be drawn from this study, therefore, is that most of the participants could decode, but not accurately enough to read fluently because of poor letter sound knowledge. The inaccurate slow reading hindered their capacity to attend to comprehension. Spaull et al. (2020:14) mention that ‘only once reading becomes relatively fast and accurate, do other variables such as vocabulary knowledge, inferencing abilities, text genre, and background knowledge account for differences in reading comprehension’. However, these researchers state that studies showing high oral reading fluency and high reading comprehension in African languages are yet to be conducted to corroborate their proposition. The outcome of this study thus calls for the enhancement of Sesotho Foundation Phase curriculum for:

Also, the curriculum for initial teacher training should include the pedagogical content knowledge of the components of reading as encapsulated in the AVR generally, and the relationship between decoding or word recognition, reading fluency and comprehension.

This study analysed the relationship between letter sound knowledge, oral reading fluency and oral reading comprehension of Sesotho Grade 4 learners in the bigger quest to identify possible contributing factors of poor reading comprehension of written texts. The results are clear on the importance of both simple and complex letter sounds for decoding texts accurately, and thus fluently, to facilitate comprehension. This revelation highlights that competency in letter sounds and function words cannot be disregarded as possible causes of reading comprehension problems and must, thus, be considered for research, reading instruction and intervention in both the foundation and the intermediate phases. It is, however, important to consider that the study was conducted during the COVID-19 period where learning loses were incurred as schools tried to adhere to protocols for preventing the spread of the virus. This cohort of learners who lost the chance to build and consolidate the necessary skills and knowledge to read with comprehension may have somewhat recovered or continue to carry this deficit throughout their schooling life depending on the support received through structured interventions.

The implications of this study therefore may be that going forward, foundation phase learners’ reading data, especially that for Grade 3 learners, must be taken into consideration in relation to their performance in the PIRLS and other reading assessments. For instance, the EGRA results, which assess the learners’ letter sound knowledge and reading fluency in the foundation phase, could be correlated to PIRLS results to determine learners whose reading comprehension may have been affected by a deficit of some foundational skills. The reading gaps that these learners incurred because of the pandemic should have been systematically addressed. Wanzek et al. (2010) state that reading gaps persist and widen if intervention is not timeously implemented to close them. Although various interventions were initiated by the Department of Education post the pandemic, it is not clear whether they were specifically dedicated to reading. The absence of systematic reading interventions may mean this cohort of learners will continue with a reading deficit throughout their schooling careers and probably adulthood.